Decagon Product Foundations

Enabling an AI agent to take action, like processing refunds, updating account details, or troubleshooting issues, requires structured workflows. That’s where Agent Operating Procedures (AOPs) come in. They’re the playbooks that allow Decagon agents to resolve complex, multi-turn conversations reliably at scale.

An AOP is a step-by-step set of natural language instructions backed by code. This design combines human-like adaptability with the control needed for production systems. Decagon built AOPs on three principles: they’re fast to create and iterate on, flexible yet durable, and equipped with guardrails to keep sensitive operations safe.

Creating an AOP starts with a descriptive name and entry conditions, the triggers that tell the agent when to use that workflow. From there, AOP Copilot can generate a first draft from plain English or an existing standard operating procedure, saving time and letting teams iterate quickly. Each workflow is structured as a series of nodes, moving step by step until the process is complete.

To learn more about AOP node types and what makes a good AOP, watch the deep dive video:

AOPs become even more powerful when extended with tools and metadata. Tools are custom logic like API calls, data lookups, or calculations that the AI agent can run in real time. Metadata provides extra context, such as location, role, or product tier, allowing workflows to adapt based on customer profile. Together, they connect the agent to live systems and personalize responses.

To support debugging and continuous improvement, Trace View shows exactly how an AOP executes — which steps were triggered, skipped, or paused.

In summary, AOPs give CX teams both control and flexibility: structured enough to be reliable, but open enough for natural, conversational AI.